[PATCH 2/2] md: batch flush requests.

[NeilBrown <neilb@xxxxxxxx>]

Currently if many flush requests are submitted to an md device is quick
succession, they are serialized and can take a long to process them all.
We don't really need to call flush all those times - a single flush call
can satisfy all requests submitted before it started.
So keep track of when the current flush started and when it finished,
allow any pending flush that was requested before the flush started
to complete without waiting any more.

Signed-off-by: NeilBrown <neilb@xxxxxxxx>
---
 drivers/md/md.c |   27 +++++++++++++++++++++++----
 drivers/md/md.h |    3 +++
 2 files changed, 26 insertions(+), 4 deletions(-)

diff --git a/drivers/md/md.c b/drivers/md/md.c
index 021e522001c1..d0f688399a56 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -427,6 +427,7 @@ static void submit_flushes(struct work_struct *ws)
 	struct mddev *mddev = container_of(ws, struct mddev, flush_work);
 	struct md_rdev *rdev;
 
+	mddev->start_flush = ktime_get_boottime();
 	INIT_WORK(&mddev->flush_work, md_submit_flush_data);
 	atomic_set(&mddev->flush_pending, 1);
 	rcu_read_lock();
@@ -467,6 +468,7 @@ static void md_submit_flush_data(struct work_struct *ws)
 	 * could wait for this and below md_handle_request could wait for those
 	 * bios because of suspend check
 	 */
+	mddev->last_flush = mddev->start_flush;
 	mddev->flush_bio = NULL;
 	wake_up(&mddev->sb_wait);
 
@@ -481,15 +483,32 @@ static void md_submit_flush_data(struct work_struct *ws)
 
 void md_flush_request(struct mddev *mddev, struct bio *bio)
 {
+	ktime_t start = ktime_get_boottime();
 	spin_lock_irq(&mddev->lock);
 	wait_event_lock_irq(mddev->sb_wait,
-			    !mddev->flush_bio,
+			    !mddev->flush_bio ||
+			    ktime_after(mddev->last_flush, start),
 			    mddev->lock);
-	mddev->flush_bio = bio;
+	if (!ktime_after(mddev->last_flush, start)) {
+		WARN_ON(mddev->flush_bio);
+		mddev->flush_bio = bio;
+		bio = NULL;
+	}
 	spin_unlock_irq(&mddev->lock);
 
-	INIT_WORK(&mddev->flush_work, submit_flushes);
-	queue_work(md_wq, &mddev->flush_work);
+	if (!bio) {
+		INIT_WORK(&mddev->flush_work, submit_flushes);
+		queue_work(md_wq, &mddev->flush_work);
+	} else {
+		/* flush was performed for some other bio while we waited. */
+		if (bio->bi_iter.bi_size == 0)
+			/* an empty barrier - all done */
+			bio_endio(bio);
+		else {
+			bio->bi_opf &= ~REQ_PREFLUSH;
+			mddev->pers->make_request(mddev, bio);
+		}
+	}
 }
 EXPORT_SYMBOL(md_flush_request);
 
diff --git a/drivers/md/md.h b/drivers/md/md.h
index 2deb84fa93f9..257cb4c9e22b 100644
--- a/drivers/md/md.h
+++ b/drivers/md/md.h
@@ -463,6 +463,9 @@ struct mddev {
 	 */
 	struct bio *flush_bio;
 	atomic_t flush_pending;
+	ktime_t start_flush, last_flush; /* last_flush is when the last completed
+					  * flush was started.
+					  */
 	struct work_struct flush_work;
 	struct work_struct event_work;	/* used by dm to report failure event */
 	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);